This had to do with leaking privacy sensitive information such as visited links by reading back pixels from a rendered <a>. They fixed this with a concept they call "read-back-allowed rendering"
In the context of knee bending: I was creating a simple walking animation with gemini 3 pro and could animate all the poses instead of one, left leg bend right leg straight. Right leg bend it had no problem but this one it just did not seem to understand.
As a proof of concept I've been working on a Jev style vision model that can rapidly answer boolean and choice questions based on images (See [0] but not the point of this comment). One of the reasons I wanted to make this concept is because I was also thinking about exactly what the parent comment suggests. I just haven't gotten around to getting (3d printing) a robot arm yet :)
Nice work! Vision on the qwen models works very good. Jev is a really neat idea. I tried replicating jev style with qwen models using max 1 token output, which works great. How does this compare to that in terms of speed?
The problem is that with max 1 token it is not 100% guaranteed that it will be a yes/no or a score. This model (and engine) solves that by taking the probability of the yes/no token. For 1 question it is faster but not a whole lot, the power comes from being able to ask more then 1 questions for merely a few ms extra per question.
E.g. 1 question would be 300ms but 30 questions would be 360ms total. Most of the time comes from the image decoding.
I've seen on X that someone made a self driving car driven by Jev. I didn't check in details what it was about. I suppose you could give it some finite set of answers. Just wondering if someone tried it.
With a car I can imagine asking questions like "am I on the left side of the lane", "am I closer then 10 meter to the next car" which feel like questions the current models could answer. I see models understanding the calibration of the robot. Maybe you could have a jev style model make decisions like "move left", "activate gripper" and do the inverse kinematics in a different system.
I did some energy price forecasting and my neural network learned the morning evening patterns just from historical data, it worked great. Then I tried adding more information to this like wind and solar (and many more). None of these external factors made any difference while being very important for energy prices. It seems like the market already incorporated all the factors in the price (as they should).
The takeaway might be that historic data of market data might often be enough to make a reasonable prediction. Only external data that nobody else has (used) can make your prediction better then the market.
Making a prediction with same accuracy as the market: easy
Making a prediction with more accuracy then the market: very hard
> The takeaway might be that historic data of market data might often be enough to make a reasonable prediction. Only external data that nobody else has (used) can make your prediction better then the market.
that's basically the definition of an efficient market. demand forecasting and insuring against price shifts is the actually useful thing that the commodities futures markets do.
also a fundamental difference from other time series prediction problems. there are all sorts of weird dynamics that go on before one biosignal effects another, or one metric predicts a failure, where a reasonably efficient market price reacts quickly to well known exogenous factors.
Cool app! I also had some ideas a while ago that I think the current breed of aim trainers are still missing.
In-game practice like the Range, DM, and Team DM still feels like the gold standard to me. I generally only use aim trainers when they offer a noticeably higher return per minute, rather than just teaching me how to get good at aim trainer specific mechanics.
Here are a couple of things I think an aim trainer should have:
- Dynamic difficulty: I feel it works best when trying to keep accuracy relatively high, but fluctuating the target accuracy threshold over time and adjusting difficulty dynamically to match it. That keeps you challenged without breaking form.
- Dynamic scenarios: Instead of picking scenarios manually, I think a trainer should ideally detect what you suck at and launch those directly.
- Habit coaching: It would be awesome if a coach flagged bad habits in real time—like clicking too fast (low accuracy), doing a slow tracking movement instead of a sharp flick (mircoflicks), or using tracking when click timing is needed.
- Biomechanics: Finding the exact "perfect" sensitivity feels less important to me than ensuring your arm and wrist are working together properly, altough this might depend on your aiming style.
- Simplicity: A lot of trainers overwhelm you with numbers and menus. I image a simple "click here to start and follow feedback" workflow with nothing else for the user to learn. Seeing a bunch of numbers is less important to me then just getting better (maybe with a final score).
Valorant-specific focus: Trainers often feel too generic for tactical shooters like Valorant, where angle holding, target reading, and micro-corrections dominate.
- In this tool, difficulty scales dynamically while playing (the parameter in parentheses changes on the fly):
- Angle Hold (enemy speed increases): Practicing holding angles against multiple targets. Holding angles is crucial in Valorant (just look at how Primmie plays).
- Follow (standstill window decreases): Forces active mouse correction and reading the target before shooting. I tend to click too fast, so this forces me to slow down and micro-correct.
- Jiggle (jiggle speed increases): Targets perform tight jiggles. I find players who spam small jiggles really hard to hit, so this directly targets that weakness.
- Micro Flicks (snap distance increases): Micro flicks are easy enough to train in the Range, so I mostly use this mode when I am away from my gaming PC.
- Static (target size decreases): Standard static clicking, but with dynamic scaling as targets shrink.
- Strafe (movement speed increases): Moving targets with limited hit windows. When watching pro DM VODs, most kills are on stationary targets, but hitting moving players is still vital and tricky to isolate in-game.
I am still not very good at Valorant but the aim train ideas are interesting. Let me know what you think!
I have visited factories for work and my experience is the same. There is so much stuff that could easily be automated but is not because it is too expansive for too little value to make a custom one off machine. The big high volume things will be automated but these machines will have 90% success rate and lot's of stuff that needs to be done by hand. You can search for factory tours on youtube to get an idea. Here are two videos, an Amazon warehouse and a Tesla factory. the big heavy stuff is automated but lot's of work is still done by hand.
https://www.youtube.com/shorts/-R6cBkza17khttps://www.youtube.com/watch?v=45slYC99uUg
Do you think humanoids would be a fit on assemblying the assembly line itself? To my limited knowledge a lot of setting up the factory is making sure your line works as expected with X 9s reliability. Here dexterous humanoids are this _universal_ virtual-to-physical interface and, akin to Auto Research, could run assembly line experiments autonomosuly?
Nice, I will be following your posts! I just bought a robot arm myself, the seeed studio B601DM (€1500 6+1 axis), it works great and is open source hardware as well and a bit more solid then the so101. I also opted to not use ros, I don't want to give up control by putting another framework in between. Is your plan to see whats possible right now or do you also have ideas on how to improve sota?
Oh very cool! Looks a bit like the TRLC-DK1 (I was looking at this one for a bit).
I think pushing the sota is quite hard to do solo but we'll see. Mostly I want to get back up to speed after having not done much robotics during the last 6 years. Best way for me to learn is to just do it, so here we are. We'll see how far I get (I suspect at some point compute will be the main bottleneck)
You can make some really cool stuff with css scroll animations. I used SVG paths with a scroll animated dash offset to draw an image while scrolling. Zero javascript, it feels so smooth. https://thomaswelter.nl (the background)
firefox android does not support CSS animation-timeline, and firefox desktop needs layout.css.scroll-driven-animations.enabled. This probably should not be used for any critical features.
Can only see it on chrome though =/. I switched to Safari as the lesser of two data-harvesting evils. Or rather, with an iPhone I've already chosen my overlord. I also switched to Kagi. Trying to deGoogle myself.